Alibaba launches its most powerful model Qwen 3.8-Max, claims it can match performance of ChatGPT and Claude

Another week, another powerful AI model from China. This time, Alibaba has launched Qwen 3.8-Max, calling it its biggest and most capable AI model yet. The company says it can rival ChatGPT and Claude across several key benchmarks.

In Short

• Alibaba has unveiled Qwen 3.8-Max, claiming its biggest AI model yet
• Company says Qwen 3.8-Max can rival ChatGPT and Claude in coding, reasoning and multimodal tasks.
• Alibaba is pitching Qwen3.8-Max as a lower-cost AI co-worker for developers

It feels like if you blink, you will miss another AI model coming out of China. Just days after Moonshot AI made headlines with Kimi K3, Alibaba has unveiled Qwen 3.8-Max, calling it its biggest and most powerful model in its Qwen family yet. The company says the new model can match the performance of leading US AI heavyweights such as OpenAI's ChatGPT and Anthropic's Claude in coding, reasoning and multimodal tasks, while offering lower pricing for developers.

Alibaba’s Qwen 3.8-Max is designed for complex reasoning, autonomous coding, multimodal understanding and long-running AI agent tasks. The company also announced that the model's open weights will be released next week, alongside a smaller Qwen 3.8-27B version, allowing developers to download and fine-tune both models.

The launch of Qwen 3.8-Max is the latest in a string of powerful AI releases from China this year. Over the past few month alone, companies including Moonshot AI, DeepSeek, ByteDance, MiniMax and Z.ai have introduced increasingly powerful models as they race to challenge US AI leaders. Just days after Moonshot AI unveiled Kimi K3, Alibaba is now taking its shot with Qwen 3.8-Max.

Inside Alibaba’s Qwen 3.8-Max

Qwen 3.8-Max is built on 2.4 trillion parameters and supports a context window of up to one million tokens. According to the company this allows it to process lengthy documents, large code repositories, images and videos in a single prompt. Developers can access the model through Alibaba Cloud's Model Studio APIs, while enterprise customers can use it through QwenWork, Alibaba's AI workplace assistant.

Alibaba also highlighted the model's autonomous capabilities. According to the company, Qwen 3.8-Max can independently handle software engineering projects, perform long-horizon planning and complete coding tasks with minimal human supervision. In one internal demonstration, Alibaba said the model successfully completed a software engineering project over 16 days, handling everything from planning and coding to debugging and optimisation.

Beyond coding, the company says Qwen 3.8-Max can analyse documents, images and videos to build searchable knowledge bases, recreate software applications from screenshots, generate interactive educational content and even convert two-dimensional floor plans into 3D visualisations.

How Qwen 3.8-Max is more efficient?

According to Alibaba, one of the things that makes Qwen 3.8-Max more efficient is that it doesn't use all 2.4 trillion parameters every time a user asks it something. Instead, it activates only around 95 billion parameters when needed, helping it deliver faster responses while keeping computing costs down.

Another Chinese frontier AI model at a lower price

Alibaba is also taking on US AI companies on pricing. The company has priced Qwen 3.8-Max at US$2 per one million input tokens and US$6 per one million output tokens, making it one of the more competitively priced frontier AI models available today. It also charges US$0.25 per one million tokens for implicit caching.

Meanwhile, Alibaba claims Qwen3.8-Max performs as well as, or even better than, some of the leading AI models from OpenAI, Anthropic and Google across a range of industry benchmarks. According to the company, the model performs strongly in coding, software engineering, research, long-video understanding and computer-use tasks.

In addition Qwen3.8-Max is also said to be the highest-ranked Chinese text model on Arena.AI and ranks second globally for visual understanding, behind only Anthropic's Claude Fable 5. That said, these benchmark results were shared by Alibaba itself and have not yet been independently verified.

- Ends